Arcee CTO argues against banning Chinese AI models, citing security parity with open-source software

Lucas Atkins, chief technology officer of US open-source AI lab Arcee, has publicly argued against banning Chinese open-weight artificial intelligence models, asserting they pose no greater security risk than standard open-source software. This stance emerges amidst intensifying debate over the growing capability and popularity of Chinese AI among US enterprises, with reports suggesting the Trump administration may consider restrictions. While proprietary model makers such as OpenAI and Anthropic express concern, Atkins contends that fears of remote control or hidden backdoors are unfounded, noting that models cannot be remotely controlled once deployed. He advocates for US enterprises to foster a domestic open ecosystem and improve their own models, highlighting that Arcee itself benefits from building upon research advancements made by Chinese developers.

The debate has intensified as Chinese open-weight models, such as Moonshot AI’s Kimi K3 and Alibaba’s Qwen, gain traction for offering inference at a fraction of the token cost of closed-source models from large US labs. Atkins argues that the fear of these models being a vector for hackers is misplaced, comparing them to standard open-source software. He notes that while the models are "open weight" rather than fully open-source, the source code is largely visible and reviewable on platforms like Hugging Face, although training methods and data remain private.

Atkins described the scenario of a model containing a hidden backdoor that triggers only under specific conditions as theoretically possible but requiring "acrobatic feats" to accomplish, with slim odds of success. He explained that there is no mechanism for entities like Arcee or Alibaba to access a model once it is running in a customer’s environment. Large organisations can mitigate risks through security testing, inspection, post-training for specific uses, and examining areas like bias, toxicity, and hallucinations before deploying the models.

Despite the political chatter, Atkins suggests that US enterprises should focus on fostering a domestic open ecosystem and improving their own models rather than restricting Chinese technology. He noted that Arcee benefits from the research advancements made by Chinese developers, as the lab can learn from their work and build upon it. This reciprocal relationship allows both sides to advance, with Atkins stating that the best way to compete is to release a model that is better, rather than relying on bans.

The CTO emphasised that enterprises are increasingly building AI applications to be model-agnostic, using multiple models to avoid lock-in. Even if Chinese models offer the best price-performance ratio today, this flexibility ensures that companies are not permanently tied to a single provider. Atkins concluded that the conversation should shift from banning foreign technology to how the US can create a robust, competitive open ecosystem that gives domestic developers something meaningful to talk about.